It would be easy to believe that nothing much has changed for years at Fetcham Grove, the home of Leatherhead FC. Trees line all four sides of the ground, the facilities are basic and the 816 spectators who turned up for their Bank Holiday Monday local derby with Dorking Wanderers clearly like it that way.
But first impressions can often be deceptive. Thanks to a partnership with IBM Watson, Leatherhead are at the cutting edge when it comes to using artificial intelligence (AI) in football. And the nascent relationship has already had a significant impact at the club.
Leatherhead play in the Bostik League Premier Division, in the seventh tier of the English football pyramid, and even at this level, when it comes to haves and have-nots, Leatherhead have generally been in the latter category. Footballers at weekends, the playing squad is made up of delivery drivers, car salesmen and shop assistants.
But while money may be tight in this part of Surrey commuter belt, Leatherhead do have one huge resource their opponents do not.
Last summer the club was approached by IBM to take part in a project using AI technology to improve their performance on the pitch. At the top end of football, clubs spend millions on experts using technology and data analysis. But by enabling Leatherhead to harness Watson, IBM wanted to show that such things are not necessarily out of reach to everyone else and set out by working at a lower level.
IBM Watson started life as a research project, aimed at building AI capable of beating the very best players on the US television game show Jeopardy. In the past decade it has evolved into different services, including the capability to convert natural language into understanding and then into the ability to give answers.
At Leatherhead, each game is filmed, the matches are then tagged and the statistics can be broken down from every aspect of the game – from individual players, to set-pieces, goals, formations and even individual passes and runs.
And when you are on the pitch for Leatherhead, there is no hiding place. Even if a mistake gets missed by those on the touchline, AI records everything and reports back.
Nikki Bull, the Leatherhead manager, was initially sceptical but soon won over.
Speaking after the match against league leaders Dorking Wanderers last week, Bull reflected on an innovative season that has produced some promising results for the club.
“At the initial meeting I was thinking, we are going to get the games recorded for us and even if that was all we got, that was a bonus,” he said.
“But IBM Watson is so much more. You can ask it about a certain player and all his passes for that game come up.
“If you are on the sideline telling them ‘you are having too many touches’ you can now show them. Things might crop up in your mind and you can check them. If you think your striker has been quiet, you can look up how many shots he has had, is it less, is it more, and from where.
“As much as we don’t want to admit it, we as coaches do have biases. When you have Watson send us a report, it is totally unbiased, it doesn’t know the players, the report comes off what it sees, not emotion, what has worked, what hasn’t. It’s been a big help.”
Steady improvement with IBM Watson (season 2018/19)
When Bull and Martin McCarthy took over as manager and assistant manager last summer, the club had only two players on their books. A shortage of cash at this level means a big turnover of players – 38 turned up for a first open training session, of whom only three stayed through the season. In total, 39 players made their Leatherhead debuts this season.
The better players tend to get poached by the richer clubs, meaning a constant need to work with less established, less experienced players. They often need all the help they can get.
After a slow start to the campaign, Leatherhead went on a fine run, with the coaching team using Watson to highlight with players what was working and what wasn’t. It helped with team selections, it helped with set-pieces, and the coaches even changed the team’s formation at one point after Watson showed what was most successful as the club rose up the league and put themselves in the play-off picture.
“Travis Gregory was a player for whom it was really good early in the season,” Bull said. “If he took one player on, his completion rate was near 100 per cent, but if he took on too many he kept losing the ball. So the penny dropped. He would take on one player and lay it off. His game improved.”
Gregory, a winger, was on the books at Chelsea and Glasgow Rangers. At 24, he is a crowd favourite at Leatherhead and still young enough to hope for a future playing in the EFL. He embraced the new technology.
“It can single every player out,” he said. “It can go over every tiny detail, which is a benefit to us. Players might not want to see things when they have done badly, but they need to know what they have done wrong to correct it.
“You can get found out. It happened a couple of times to me, where I should have been covering at the back post but the opposition have scored. It’s all about learning and coping with it. If they had come to me with this at 18, I might have shied away from it, but it has helped me a lot.”
McCarthy says the system led to changes in the way the players train and prepare. “It has been a great tool for us,” he said. “We can show players the difference in completion percentage if they pass under or over 30 metres, the difference if they go past players on the right or the left. We can go in the dressing room lambast the players, say you didn’t do this or that, and be completely wrong. Our eyes can deceive us.
“We will sit down and see patterns. We have seen it develop. The players have taken it on and that is why we have had a successful season.”
The system not only works using previous footage, it also does a job researching the opposition. At a level where scouting infrastructure is basic, Watson Discovery has the ability to dig around in any available source for information.
“You can train Watson to understand domain-specific terminology – like free-kick, set-piece, corner – and it can then read through Twitter timelines and match reports and produce summary reports,” said Joe Pavitt, master inventor at IBM who has been instrumental in the Leatherhead project. “It means they can ask how other teams are going to play and what players they need to look out for.”
The 3-0 defeat by Dorking, the runaway league champions, cost Leatherhead their final chance of making the play-offs. It underlined the biggest points of success at all levels are that the team with the best players tend to be the most successful and the teams with the most money tend to be the teams with the best players.
For IBM the project with Leatherhead has shown that the technology works and is easy to understand, since despite the gulf in finances available to Leatherhead and the likes of Dorking, the gap between them has narrowed. And Pavitt believes it could be adapted for almost any sport, enabling coaches and players to get a much better understanding of sports data than ever before.
“When we originally designed it, we only planned on interfacing with the coaches and whatever they used it for they would work back into training sessions and matches,” Pavitt said. “We were pleasantly surprised that it was so easy to use and that they have given players access to it. The players have embraced it and found it very useful.
“It’s a common misconception of AI that it is very complicated. People think that it is sci-fi or it will require a lot of technical expertise to use and benefit from.
“If we can get the coaches and players to come in and interact with Watson without really thinking about it, we can show that you don’t have to be an AI expert to use it. Anyone can pick it up and benefit from it.”
Leatherhead have certainly felt the benefits, rising up the league and only narrowly missing out on the play-offs at the end of the season.
For more information, ibm.com/watson